exploring-in-parallel

You explore a codebase in parallel by launching multiple subagents simultaneously to run broad coverage searches. This collates cross-file results into a unified answer, significantly reducing wall-clock time.

Investigating a bug without sequential scanning requires launching parallel subagents to query multiple codebase areas and tooling outputs at once. This structured synthesis reconciles the findings into a unified answer with references.

A parallel subagent for codebase research is an autonomous task payload defined by a subagent_type, description, and prompt. It executes cross-file exploration concurrently with other subagents to gather multi-source context rapidly.

Yes, you can run multiple subagents simultaneously for cross-file exploration. You craft a single message containing structured Task payloads for each subagent, enabling them to execute research questions concurrently.

The best way to structure task payloads for parallel codebase exploration is to define each payload with subagent_type, description, and prompt fields. This structure allows each subagent to execute its specific research task independently.

You should avoid using parallel subagents for debugging when tasks are strictly sequential or highly interdependent. Parallel exploration is designed for broad, independent cross-file discovery rather than linear, state-dependent execution paths.